空气污染指数的前馈神经网络预测方法  

A feedforward neural network modeling method with improved learning rates for API prediction

作　　者：韩广[1] 徐璐[1] 孙晓云[1] 卢兆楠 田家辉 

机构地区：[1]石家庄铁道大学电气与电子工程学院,河北石家庄050000

出　　处：《计算机与应用化学》2016年第2期147-151,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(51274144)

摘　　要：针对石家庄空气污染指数(Air Pollution Index,简称API)的预警问题,本文提出了一种基于前馈神经网络(Feedforward Neural Network,简称FNN)的预测方法。基于三层FNN逼近任意非线性函数的特性,建立了API的预测模型;同时,针对数据变化较大的问题,采用改进学习率的梯度下降算法,对预测模型进行离线训练。选用石家庄封龙山2014年10月的空气质量数据作为研究对象,仿真实验表明,基于FNN的预测模型可以有效地对API进行预警,且预测精度可达87.94%。For the prediction and early warning of Shijiazhuang Air Pollution Index （API）, this paper proposes a method for the prediction of API via Feedforward Neural Network（FNN）. Based on three-layer FNN＇s ability in approximating to any nonlinear functions, we establish a model of API prediction; at the same time, for the problem of the various changes of data, we adopt the gradient descent algorithm with improved learning rate for the offline training of the prediction model. We selected air quality data of Shijiazhuang Fenglong Mountain in October, 2014 as object. Simulation experiments show the model based on FNN with improved learning rate can effectively make the early warning and forecast of API, and the prediction accuracy achieves 87.94%.

关 键 词：前馈神经网络 空气污染指数 预测 学习率 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]